Evaluation of intonation features on emphasized Malay words / Syazwani Nasaruddin
Detection of emphasized has become one active research in speech processing. Nowadays, in automatic speech recognition, listeners not only want to be able automatically understand which words the speakers have said but also how the speakers deliver the speech. This paper presents detection of emphas...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2017
|
| Subjects: | |
| Online Access: | https://ir.uitm.edu.my/id/eprint/98181/1/98181.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Detection of emphasized has become one active research in speech processing. Nowadays, in automatic speech recognition, listeners not only want to be able automatically understand which words the speakers have said but also how the speakers deliver the speech. This paper presents detection of emphasis in Malay sentences. Thus, the purpose of this research is consisting, to construct emphasized Malay speech, to evaluate pitch and intensity features on Malay words and lastly to classify the Malay words using intonation features. First step to construct emphasized Malay speech is by listening to 96 sentences and 314 words by 2 different speakers. In this step, user hear the speech and used the hand labelling method discuss in chapter 2. User listened to the speech audio and mark the words that the user finds the speaker voice is high. If the voice of speaker is high, user classified the word as emphasized words. This step is to determine the emphasized and non-emphasized words. All the features of emphasized words detect by user is extracted. After detection of emphasized and non-emphasized words for speakerl and speaker2, intensity and pitch features need to be evaluated. Intensity and pitch features that need to be evaluated are minimum value of pitch and intensity, maximum value of pitch and intensity and mean value of pitch and intensity. In this research, all 314 words from 2 different male speakers' speeches are clustered manually based on features of intensity and pitch pattern. The pattern of the features such as intensity and pitch are observed and clustered to find out how many patterns can be identified from all the words. Result from the intensity and pitch features can classify Malay phrases. Activities in this section are, for testing part, 314 words from 2 different speakers are evaluated by using clustering method. WEKA is a set of machine learning algorithm for data mining task. The data are clustered to determine emphasized and non-emphasized words. From the research done shows that he highest percentage in detecting emphasized words in Malay words is by combination of all features such as intensity and pitch features. |
|---|